Download Comp

Public alpha builds for Windows, macOS and Linux, plus the optional local AI Worker used by AutoMatte.

Comp application

Comp is currently available as an early public alpha for evaluation and testing. The current public alpha is Comp 0.1.2, released on August 2, 2026.

Windows x64

Extract the ZIP before running Comp. The current build is unsigned, so Windows may display a security warning — verify the checksum before allowing it to run.

Download ZIP · 51.9 MB

SHA-256 checksums

macOS arm64

The DMG is recommended for normal installation. The ZIP is provided as an alternative package.

Download DMG · 78.5 MB

Alternative: ZIP · 68.8 MB

SHA-256 checksums

Linux x86_64

The AppImage is recommended for most users. The tar.gz archive is provided as an alternative package.

Download AppImage · 76.8 MB

Alternative: tar.gz · 83.2 MB

SHA-256 checksums

Alpha software. Please keep backups of important files and use the application at your own discretion. Your feedback helps shape the direction of future development.
Linux baseline. The Linux build is packaged and tested on Rocky Linux 10.2. It may also work on other recent x86_64 distributions, but they have not yet been tested systematically.

Optional: Comp AI Worker Service

Comp can be used as a standalone compositor — you do not need the AI Worker for standard compositing. The Worker is a separate command-line component used only by the optional AI modules: SAM3 (prompted image segmentation) and MatAnyone2 (mask-edge refinement). Current Worker release: v0.1.2.

Windows — CPU / CUDA

Single package covering both CPU and CUDA targets.

Download ZIP · 1.71 GB

SHA-256 checksums

macOS — CPU / MPS

DMG recommended; ZIP provided as an alternative package.

Download DMG · 47.8 MB

Alternative: ZIP · 39.9 MB

SHA-256 checksums

View the Worker v0.1.2 release page →

Model weights are not included. Neural-network model files must be downloaded separately and placed in the folders shown below. If Hugging Face asks you to sign in or accept model conditions, complete that step first. Third-party model files have their own licenses and access conditions. MatAnyone2 is distributed under the NTU S-Lab License 1.0; check the upstream license before using its weights in your workflow. MatAnyone2 upstream →

Required model files

ModuleDownloadDestinationFinal filename
MatAnyone2 mat2_bundle.pt <worker-root>/family_bundles/mat2/models/ mat2_bundle.pt
SAM3 tokenizer bpe_simple_vocab_16e6.txt.gz <worker-root>/family_bundles/sam3/assets/ bpe_simple_vocab_16e6.txt.gz
SAM3 model sam3_bundle.pt <worker-root>/family_bundles/sam3/models/ sam3_bundle.pt
<worker-installation-root>/
└── family_bundles/
    ├── mat2/
    │   └── models/
    │       └── mat2_bundle.pt
    └── sam3/
        ├── assets/
        │   └── bpe_simple_vocab_16e6.txt.gz
        └── models/
            └── sam3_bundle.pt

Starting the Worker

Replace the placeholder paths and TARGET. Use cuda for an NVIDIA GPU, mps for Apple GPU acceleration, or cpu. Binding to 127.0.0.1 keeps the Worker local to the same machine.

macOS / Linux shell

"/path/to/comp-worker" serve \
  --family-bundle "mat2=<worker-root>/family_bundles/mat2" \
  --family-bundle "sam3=<worker-root>/family_bundles/sam3" \
  --target TARGET \
  --bind-host 127.0.0.1 \
  --port 19282 \
  --temp-dir "<path-to-temp-directory>/comp-worker" \
  --cpu-threads 0 \
  --precision fp32 \
  --verbose

Windows PowerShell

& "C:\path\to\comp-worker.exe" serve `
  --family-bundle "mat2=C:\path\to\worker\family_bundles\mat2" `
  --family-bundle "sam3=C:\path\to\worker\family_bundles\sam3" `
  --target TARGET `
  --bind-host 127.0.0.1 `
  --port 19282 `
  --temp-dir "C:\path\to\temp\comp-worker" `
  --cpu-threads 0 `
  --precision fp32 `
  --verbose

For a reduced-memory test, add --max-propagation-history-depth 2 and --max-mem-frames 2.

Community and support

Feedback from early testers directly shapes priorities for future releases. Use the community channels for bug reports, workflow feedback and build notes.

💬

Discord

Join the community to share feedback and bug reports.

discord.gg/9vgz4Xv655 →

Buy Me a Coffee

Support ongoing development of Comp.

buymeacoffee.com/yuri.r →

in

LinkedIn

Follow updates from the developer, Yuri Radaev.

linkedin.com/in/yuri-radaev →

Try the alpha and shape what's next

Windows, macOS and Linux builds are available now. Report issues and share your experience on Discord.

Browse all releases